git.sesse.net Git - bcachefs-tools-debian/blob - libbcachefs/error.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include "bcachefs.h"
   3 #include "error.h"
   4 #include "io.h"
   5 #include "super.h"
   6
   7 #define FSCK_ERR_RATELIMIT_NR   10
   8
   9 bool bch2_inconsistent_error(struct bch_fs *c)
  10 {
  11         set_bit(BCH_FS_ERROR, &c->flags);
  12
  13         switch (c->opts.errors) {
  14         case BCH_ON_ERROR_continue:
  15                 return false;
  16         case BCH_ON_ERROR_ro:
  17                 if (bch2_fs_emergency_read_only(c))
  18                         bch_err(c, "inconsistency detected - emergency read only");
  19                 return true;
  20         case BCH_ON_ERROR_panic:
  21                 panic(bch2_fmt(c, "panic after error"));
  22                 return true;
  23         default:
  24                 BUG();
  25         }
  26 }
  27
  28 void bch2_topology_error(struct bch_fs *c)
  29 {
  30         set_bit(BCH_FS_TOPOLOGY_ERROR, &c->flags);
  31         if (test_bit(BCH_FS_INITIAL_GC_DONE, &c->flags))
  32                 bch2_inconsistent_error(c);
  33 }
  34
  35 void bch2_fatal_error(struct bch_fs *c)
  36 {
  37         if (bch2_fs_emergency_read_only(c))
  38                 bch_err(c, "fatal error - emergency read only");
  39 }
  40
  41 void bch2_io_error_work(struct work_struct *work)
  42 {
  43         struct bch_dev *ca = container_of(work, struct bch_dev, io_error_work);
  44         struct bch_fs *c = ca->fs;
  45         bool dev;
  46
  47         down_write(&c->state_lock);
  48         dev = bch2_dev_state_allowed(c, ca, BCH_MEMBER_STATE_ro,
  49                                     BCH_FORCE_IF_DEGRADED);
  50         if (dev
  51             ? __bch2_dev_set_state(c, ca, BCH_MEMBER_STATE_ro,
  52                                   BCH_FORCE_IF_DEGRADED)
  53             : bch2_fs_emergency_read_only(c))
  54                 bch_err(ca,
  55                         "too many IO errors, setting %s RO",
  56                         dev ? "device" : "filesystem");
  57         up_write(&c->state_lock);
  58 }
  59
  60 void bch2_io_error(struct bch_dev *ca)
  61 {
  62         //queue_work(system_long_wq, &ca->io_error_work);
  63 }
  64
  65 #ifdef __KERNEL__
  66 #define ask_yn()        false
  67 #else
  68 #include "tools-util.h"
  69 #endif
  70
  71 static struct fsck_err_state *fsck_err_get(struct bch_fs *c, const char *fmt)
  72 {
  73         struct fsck_err_state *s;
  74
  75         if (test_bit(BCH_FS_FSCK_DONE, &c->flags))
  76                 return NULL;
  77
  78         list_for_each_entry(s, &c->fsck_errors, list)
  79                 if (s->fmt == fmt) {
  80                         /*
  81                          * move it to the head of the list: repeated fsck errors
  82                          * are common
  83                          */
  84                         list_move(&s->list, &c->fsck_errors);
  85                         return s;
  86                 }
  87
  88         s = kzalloc(sizeof(*s), GFP_NOFS);
  89         if (!s) {
  90                 if (!c->fsck_alloc_err)
  91                         bch_err(c, "kmalloc err, cannot ratelimit fsck errs");
  92                 c->fsck_alloc_err = true;
  93                 return NULL;
  94         }
  95
  96         INIT_LIST_HEAD(&s->list);
  97         s->fmt = fmt;
  98         s->buf = PRINTBUF;
  99         list_add(&s->list, &c->fsck_errors);
 100         return s;
 101 }
 102
 103 int bch2_fsck_err(struct bch_fs *c, unsigned flags, const char *fmt, ...)
 104 {
 105         struct fsck_err_state *s = NULL;
 106         va_list args;
 107         bool print = true, suppressing = false, inconsistent = false;
 108         struct printbuf buf = PRINTBUF, *out = &buf;
 109         int ret = -BCH_ERR_fsck_ignore;
 110
 111         mutex_lock(&c->fsck_error_lock);
 112         s = fsck_err_get(c, fmt);
 113         if (s) {
 114                 if (c->opts.ratelimit_errors &&
 115                     !(flags & FSCK_NO_RATELIMIT) &&
 116                     s->nr >= FSCK_ERR_RATELIMIT_NR) {
 117                         if (s->nr == FSCK_ERR_RATELIMIT_NR)
 118                                 suppressing = true;
 119                         else
 120                                 print = false;
 121                 }
 122
 123                 printbuf_reset(&s->buf);
 124                 out = &s->buf;
 125                 s->nr++;
 126         }
 127
 128         if (!strncmp(fmt, "bcachefs:", 9))
 129                 prt_printf(out, bch2_log_msg(c, ""));
 130
 131         va_start(args, fmt);
 132         prt_vprintf(out, fmt, args);
 133         va_end(args);
 134
 135         if (test_bit(BCH_FS_FSCK_DONE, &c->flags)) {
 136                 if (c->opts.errors != BCH_ON_ERROR_continue ||
 137                     !(flags & (FSCK_CAN_FIX|FSCK_CAN_IGNORE))) {
 138                         prt_str(out, ", shutting down");
 139                         inconsistent = true;
 140                         ret = -BCH_ERR_fsck_errors_not_fixed;
 141                 } else if (flags & FSCK_CAN_FIX) {
 142                         prt_str(out, ", fixing");
 143                         ret = -BCH_ERR_fsck_fix;
 144                 } else {
 145                         prt_str(out, ", continuing");
 146                         ret = -BCH_ERR_fsck_ignore;
 147                 }
 148         } else if (c->opts.fix_errors == FSCK_OPT_EXIT) {
 149                 prt_str(out, ", exiting");
 150                 ret = -BCH_ERR_fsck_errors_not_fixed;
 151         } else if (flags & FSCK_CAN_FIX) {
 152                 if (c->opts.fix_errors == FSCK_OPT_ASK) {
 153                         prt_str(out, ": fix?");
 154                         bch2_print_string_as_lines(KERN_ERR, out->buf);
 155                         print = false;
 156                         ret = ask_yn()
 157                                 ? -BCH_ERR_fsck_fix
 158                                 : -BCH_ERR_fsck_ignore;
 159                 } else if (c->opts.fix_errors == FSCK_OPT_YES ||
 160                            (c->opts.nochanges &&
 161                             !(flags & FSCK_CAN_IGNORE))) {
 162                         prt_str(out, ", fixing");
 163                         ret = -BCH_ERR_fsck_fix;
 164                 } else {
 165                         prt_str(out, ", not fixing");
 166                 }
 167         } else if (flags & FSCK_NEED_FSCK) {
 168                 prt_str(out, " (run fsck to correct)");
 169         } else {
 170                 prt_str(out, " (repair unimplemented)");
 171         }
 172
 173         if (ret == -BCH_ERR_fsck_ignore &&
 174             (c->opts.fix_errors == FSCK_OPT_EXIT ||
 175              !(flags & FSCK_CAN_IGNORE)))
 176                 ret = -BCH_ERR_fsck_errors_not_fixed;
 177
 178         if (print)
 179                 bch2_print_string_as_lines(KERN_ERR, out->buf);
 180
 181         if (!test_bit(BCH_FS_FSCK_DONE, &c->flags) &&
 182             (ret != -BCH_ERR_fsck_fix &&
 183              ret != -BCH_ERR_fsck_ignore))
 184                 bch_err(c, "Unable to continue, halting");
 185         else if (suppressing)
 186                 bch_err(c, "Ratelimiting new instances of previous error");
 187
 188         mutex_unlock(&c->fsck_error_lock);
 189
 190         printbuf_exit(&buf);
 191
 192         if (inconsistent)
 193                 bch2_inconsistent_error(c);
 194
 195         if (ret == -BCH_ERR_fsck_fix) {
 196                 set_bit(BCH_FS_ERRORS_FIXED, &c->flags);
 197         } else {
 198                 set_bit(BCH_FS_ERRORS_NOT_FIXED, &c->flags);
 199                 set_bit(BCH_FS_ERROR, &c->flags);
 200         }
 201
 202         return ret;
 203 }
 204
 205 void bch2_flush_fsck_errs(struct bch_fs *c)
 206 {
 207         struct fsck_err_state *s, *n;
 208
 209         mutex_lock(&c->fsck_error_lock);
 210
 211         list_for_each_entry_safe(s, n, &c->fsck_errors, list) {
 212                 if (s->ratelimited)
 213                         bch_err(c, "Saw %llu errors like:\n    %s", s->nr, s->buf.buf);
 214
 215                 list_del(&s->list);
 216                 printbuf_exit(&s->buf);
 217                 kfree(s);
 218         }
 219
 220         mutex_unlock(&c->fsck_error_lock);
 221 }